Best practice for mounting disks


#1

what’s the best practice in terms of mounting ssd disk (ebs, for example) for cockroachdb? Assume that I need to allocate 500GB per node, is it better to have 5x 100GB disks per node using store flags or just mount 1x 500GB?

Thanks!


(Jesse) #2

Hi @jeorge,

This is a great question, and our production reconmendations are incomplete on the topic. I’ll make sure we add more detail.

In the meantime, our basic guidance is that it’s better to use one larger disk per node rather than several smaller disks. In some cases, CockroachDB may not balance data optimally across nodes with multiple smaller disks each.

Hope that helps.

Best,
Jesse


#3

Thanks @jesse
Let’s say I have 3 node with 3 replicas with approximately 256 bytes per row and I am anticipating 1 billion records which would translate to ~ 256G, do you think 300G disk per node would be enough considering additional time-series data?

Thanks again!


(Pete Vilter) #4

Hi @jeorge,
I’m taking this thread over from @jesse.

300gb per node sounds reasonable. The exact disk usage is hard to predict, since there is compression, but there is also some extra book-keeping data associated with Raft, etc. I think those two roughly even out.

I don’t have a formula readily available for time series space usage, but a couple of our long-running test clusters with six nodes have 1.9gb of time series data, so I’d expect it to stay around that. The cluster setting documented here can be used to control time series data usage: https://www.cockroachlabs.com/docs/stable/operational-faqs.html#can-i-reduce-or-disable-the-storage-of-timeseries-data

Hope that helps.